-
-
Notifications
You must be signed in to change notification settings - Fork 164
Model descriptions
In this article, information is provided about the most relevant models from OpenAI. All models have been trained on data up until September 2021. Keep in mind that the context is provided for the entire conversation, not just one message. You can use the OpenAI tokenizer for token and character counting.
Unlike GPT-3, this model is optimized for conversational chat. GPT-3.5 has been trained using 175 billion parameters gathered from various sources. The distribution of these sources for model training is as follows:
- Web scraping: 60%
- Web text: 22%
- Books: 15%
- Wikipedia: 3%
| Model | Description | Release date | Context |
|---|---|---|---|
| gpt-3.5-turbo | The most efficient and optimized model, it is sufficient for most tasks. | November 28, 2022 | 4097 tokens |
| gpt-3.5-turbo-0301 | Improved performance and efficiency. It has a greater ability to understand and generate various dialects, respond to emotions expressed in the text. More diverse and natural responses. | March 1, 2023 | 4097 tokens |
| gpt-3.5-turbo-0613 | Improved model control and the ability to get more predictable responses. Processing time is reduced by 40%, but the quality of the responses is slightly worse. | June 13, 2023 | 4097 tokens |
| gpt-3.5-turbo-16k | The same capabilities as the standard gpt-3.5-turbo model, but with a 4x larger context. It can process approximately 20 pages of English text per query. | November 28, 2022 | 16385 tokens |
| gpt-3.5-turbo-16k-0613 | Enhanced capabilities for processing large volumes of text and improved functionality for invoking. | June 13, 2023 | 16385 tokens |
| gpt-3.5-turbo-instruct | Difference between gpt-3.5-turbo-instruct and gpt-3.5-turbo models lies in their functionality. gpt-3.5-turbo-instruct model is trained to provide direct answers or give specific instructions, while gpt-3.5-turbo is optimized for dialogues. | September 19, 2023 | 4097 tokens |
The model is capable of understanding both text and images. Creative abilities, working with code, and performing complex logical tasks have been improved. For most tasks, the difference between GPT-4 and GPT-3.5 is not noticeable. However, in more complex situations, GPT-4 is more effective.
| Model | Description | Release date | Context |
|---|---|---|---|
| gpt-4-0314 | A more powerful model than GPT-3.5. It can solve complex tasks with greater accuracy. | March 14, 2023 | 8192 tokens |
| gpt-4-0613 | Faster, the query execution time has been reduced by 20%. It has an updated and improved architecture. It represents a more advanced and enhanced version of the GPT-4 model with more accurate results, the ability to call functions, and an increased context length. | June 13, 2023 | 8192 tokens |
| gpt-4-32k-0314 | Same capabilities as the standard GPT-4 model, but with a context length increased fourfold. | March 14, 2023 | 32768 tokens |
| gpt-4-32k-0613 | It has an extended context length for better understanding of larger texts. This is useful when tackling complex programming tasks and processing large volumes of code. | June 13, 2023 | 32768 tokens |
| gpt-4v | Allows asking questions about images and using speech as input data. | September 24, 2023 | 8192 tokens |